Republican Senator Tom Cotton of Arkansas has called on the Trump administration to adopt a sweeping, multi-pronged strategy to curb American reliance on Chinese artificial intelligence models, warning that national security risks are being subordinated to commercial convenience. In a letter addressed to Commerce Secretary Howard Lutnick on Wednesday, 30 July 2025, and released publicly on Thursday, 31 July 2025, Cotton proposed tighter export controls, broader contractor restrictions, and a push to make American AI the global standard.

What Cotton's Letter Demands

Cotton's core proposal is a government-wide ban on the use of Chinese AI models by federal contractors, extending prohibitions that already exist for DeepSeek to cover additional Chinese models. He noted that the Department of War (DOW) has already barred both agency and contractor use of DeepSeek, and that several other federal agencies have blocked internal access. Cotton argued that widening this ban would send a clear signal that Washington will not trade security for cost savings.

He also called for maintaining strict export controls on advanced computing hardware and semiconductor manufacturing equipment, and for stronger government support of a competitive domestic AI ecosystem capable of rivalling Chinese offerings in both home and international markets.

The Scale of Chinese AI Penetration

Cotton cited a report estimating that nearly 80 per cent of US startups build applications on Chinese AI models — a figure he described as alarming. He pointed to specific commercial examples: Airbnb's chatbot powered by Qwen and Cursor's coding agent built on Kimi. The latter drew particular concern because, according to Cotton, defence contractors use Cursor's coding tools to build software for the government, even though he claimed Chinese models generate 'lower quality, more vulnerable code when the user is American.'

He also flagged that Microsoft was reportedly considering DeepSeek as an alternative for its Copilot Cowork platform, describing the potential move as a risk to supply-chain integrity.

Security Risks: Backdoors and Auditing Gaps

Cotton warned that backdoors may be embedded in open-weight Chinese models and that auditing provides only limited assurances. 'American companies, particularly those that do business with the US government, should never jeopardise our security for higher profits,' he wrote. Critics of open-weight model deployment have long raised similar concerns, but Cotton's letter marks one of the more direct congressional calls for binding restrictions rather than voluntary guidance.

The Broader AI Race Argument

Beyond security, Cotton framed the issue as an economic and geopolitical contest. He argued that the United States must ensure its AI models become the global standard — setting 'standards, norms, and dependencies' — and that allowing Chinese models to dominate global diffusion would cede that structural advantage. He described Communist China as pursuing an 'aggressive AI diffusion strategy' and said Washington must respond with equal ambition.

His vision spans the full capability spectrum: frontier models for defence, critical infrastructure, and advanced scientific research at the high end, and cheaper, efficient models for commercial and consumer workloads where price competitiveness matters more. 'America wins the AI race when we offer models that provide optionality,' Cotton wrote, concluding with a call for prompt action from the administration.

What Comes Next

The letter is a congressional pressure point rather than binding policy, but it arrives as the administration is already tightening AI-related export controls and reviewing supply-chain exposure to Chinese technology. Whether the Commerce Department acts on Cotton's specific recommendations — particularly the government-wide contractor ban — will be a key indicator of how aggressively Washington intends to prosecute the AI competition with Beijing in the months ahead.